Some big news, to me at least: The Bureau of Economic Analysis is now producing "regional price parities" data that allow you to compare the cost of living in one place in the US to another. The BEA news release release is here; coverage from the tax foundation here (HT the always interesting Marginal Revolution). In the past, you could see regional inflation -- changes over time -- but you couldn't compare the level of prices in different places.
The states differ widely. It is in fact as if we live in different countries with different currencies. Hawaii (116.8) vs. Mississippi (86.7) is bigger than paying in dollars vs Euros (118) Yen (times 100, 1.01) and almost as big as pounds (1.30)
The variation across city/country and across cities is even higher:
In 2014, the metropolitan area with the highest RPP was Urban Honolulu, HI (123.5). Metropolitan areas with RPPs above 120.0 also included San Jose-Sunnyvale-Santa Clara, CA (122.9), New York-Newark-Jersey City, NY-NJ-PA (122.3), Santa Cruz-Watsonville, CA (121.8), San Francisco-Oakland-Hayward, CA (121.3), and Bridgeport-Stamford-Norwalk, CT (120.4). The metropolitan area with the lowest RPP was Beckley, WV (79.7), followed by Rome, GA (80.7), Danville, IL (81.1), Morristown, TN (81.9), and Jonesboro, AR (82.0).No surprise, much of the variation is due to housing. Breaking it out, (look up your town here!)
San Francisco-Oakland-Hayward, CA
All items 121.3
Goods 108.4
Services: Rents 183.9
Services: Other 109.6
San Jose-Sunnyvale-Santa Clara, CA
All items 122.9
Goods 108.2
Services: Rents 200.7
Services: Other 109.3
Beckley, WV
All items 79.7
Goods 92
Services: Rents 52.8
Services: Other 92.5
There is still a 20% difference in the cost of goods and other services, but the variation in rents is really big. When you consider that the cost of real estate drives up other costs, its effect may be even larger: If the barbershop pays higher rent, and the barber pays higher rent, you're going to pay more for haircuts. And this is just rents. Since houses have thin rental markets, the true difference may be larger still. Also, rents are often controlled or poorly measured. I don't know how BLS deals with that.
You can see many uses for even more granular data. But since house price and rent are easy to get, you might get a good approximation by adding granular housing cost data to regional price data.
There are a lot of interesting issues here.
One question it raises is the true picture of inequality. Poor people, especially those who don't work, tend to live in low-rent areas. Relative to local prices, inequality may not be as bad as it seems. (I presume the BLS does something to adjust rents for quality of housing.)
One can also imagine that congresspeople from high price areas will soon ask for higher cost of living adjustments for benefits to their constituents.
This data ought to focus more attention on housing supply restrictions -- the main reason that rents vary so much.
It raises some puzzles too. I notice that the market for academics gives surprisingly little weight to cost of living variations. If you compare offers from a European and US university, nobody expects you to compare "100,000" in each place without converting currency. But nominal academic salaries are quite similar across chasms of cost of living. To some extent universities make it up with absurdly complex and inefficient housing subsidies, but that doesn't make much sense either. I'm curious to what extent this phenomenon occurs in other markets.
And... who knows? New data always leads to interesting new research. Kudos to the BEA for making this available.
Comments from people who know how this data is constructed, with good parts and pitfalls, are especially welcome.
Update
A colleague who knows a lot about these issues sent some useful information:
...it’s my understanding from conversations with a few people and brief reading on methodology (https://www.bea.gov/regional/pdf/RPP2015.pdf) that they are actually pretty poor measures of local prices. Essentially all of the variation comes from relatively poorly measured housing prices, almost by construction.
That’s because the only local retail price data going into the BEA indices comes from the BLS CPI data, which covers less than 30 cities (and not even on identical products across locations). They’re extrapolating from this small number of cities to all cities in the US by just taking the nearest city with CPI data and re-weighting it with local expenditures shares. So for example, there is no retail pricing data collected for Columbus, but they show up in the BEA metro area price parities. So where are they getting price data from? They just take the prices collected in Cleveland (where BLS collects data) and assume that are the same in Columbus with potentially slightly different weights in the consumption basket. So even if there is wide heterogeneity across cities in prices... this is for the most part not going to get picked up in their local price measures, since they’re imputing prices in most cities using pricing data from other cities. Since most states have either 0 or 1 BLS price collection cities, this means that close to 100% of the within-state variation in their price levels is coming from housing. So to close to a first approximation, these purchasing power indices are really just house price indices since they basically aren’t using data on local prices for anything except housing.
But the housing price data is coming from ACS with various hedonic adjustment. That is notoriously challenging, especially across locations. It’s much easier but still hard to compute house price changes across time using repeat sales indices like core logic, but the housing stock is fundamentally heterogeneous across space which puts huge standard errors on trying to construct the price for an equivalent unit of housing across space, so I take the exact numbers there with a big grain of salt.
So overall I think these indices basically just tell you that housing is more expensive in san francisco and NYC than in oklahoma, but I think their quantitative usefulness is pretty limited. I think to really measure price level differences across locations, scanner data is much more useful since we can measure identical products as well as product availability and varieties. (A weakness is that this can’t capture differences in service prices across space, but it’s hard to adjust for quality there just like for housing, even if we had a census of all service providers prices everywhere in the country). Jessie Handbury and David Weinstein’s 2014 restud paper is the best study I know of trying to take seriously measuring retail price levels across locations using that kind of data. I have no idea how it lines up with the BEA numbers.
From which I take: 1) This is very important 2) The BLS took a useful stab at it with the numbers they have but 3) understand the large limitations of the BLS numbers before you use them 4) get to work, big-data economists, on using scanner data, twitter feeds, amazon purchases, zillow, and everything else you can get your hands on, to produce 21st century granular price indices!
Update 2:
Enrico Moretti has already written a very nice paper, Real wage inequality (Also here) adjusting inequality measures for local cost of living.
At least 22% of the documented increase in college premium is accounted for by spatial differences in the cost of living.He creates local price indices. He also takes on the question whether higher prices in hot cities represent more housing -- better amenities -- or just higher prices which you have to pay in order to work high -productivity jobs.
